Add an API for determining if an object is immortal

[ZeroIntensity]

Feature or enhancement

Proposal:

Immortal objects might lead to some unexpected results for users that are (unfortunately) messing with reference count details, as seen in #127191; but currently, there's no way to determine what objects are actually immortal.

Immortality is an implementation detail, but it still does affect reference counting, and given that it's not very safe to rely on very specific reference counts, we don't want the only way to check for immortality to be via sys.getrefcount. So, given that we already have sys._is_interned, it seems fitting to add a sys._is_immortal alongside it.

If we go with this, I think it's also a good idea to note it in some of the documentation. For example, sys.getrefcount says this:

some objects are immortal and have a very high refcount that does not reflect the actual number of references.

Here, it might be a good idea to mention my proposed sys._is_immortal, so there's less ambiguity about the returned value.

cc @erlend-aasland

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

Linked PRs

• gh-128509: Add sys._is_immortal for identifying immortal objects #128510
• gh-128509: Add PyUnstable_IsImmortal for finding immortal objects #129182

[kumaraditya303]

Immortality is an implementation detail, but it still does affect reference counting, and given that it's not very safe to rely on very specific reference counts, we don't want the only way to check for immortality to be via sys.getrefcount. So, given that we already have sys._is_interned, it seems fitting to add a sys._is_immortal alongside it.

What is the use-case to check if something is immortal at python level? Adding it for consistency reason alone isn't enough as it a very specific impl detail of CPython. I would like to know the use case before exposing it at Python level.

[ZeroIntensity]

Differentiating between "a lot of references" and an immortal object (e.g. for debugging). It's not a good idea to have users rely on an object having a specific number of references to declare it immortal, because we change the immortal reference count, and objects with deferred reference counts also have "very large" reference counts. (FWIW, immortality might not even be done with a fixed refcount someday--that's not something we can deprecate at a sys.getrefcount level.)

[vstinner]

Maybe we could start by adding a C API for that? PyUnstable_IsImmortal() or even Py_IsImmortal()?

[vstinner]

@kumaraditya303:

What is the use-case to check if something is immortal at python level?

Writing unit tests I would say.

Since sys.getrefcount() is public, IMO it makes sense to add sys._is_immortal(). We should not and cannot this "implementation detail". It's not a minor detail, it does confuse a lot of people (me included!).